Politically Speaking: Ethnic Language and Audience Opinion in Southeast Asia

Abstract

Language is one of the quintessential markers of ethnicity. It allows co-ethnics to easily identify one another and underscores in-group and out-group boundaries. Recognizing this, politicians frequently employ ethnic tongues to enhance their political appeal. To what extent does this shape the opinions of their audiences? Utilizing a survey experiment, I test the impact of an ethnic tongue against that of the common political language among the Javanese in Indonesia, the Tagalog in the Philippines, and the Isan people in Thailand. The experiment demonstrates that the ethnic language has a significant impact in both Thailand and Indonesia, but there appears to be little effect of using Tagalog over Filipino English in the Philippines. The findings suggest that ethnic tongues have the potential to significantly enhance political appeals, both among dominant (Javanese) and marginal (Isan) ethnic groups, but when the ethnic group is already the linguistic hegemon (Tagalog), such effects may be limited.

Language is one of the quintessential markers of ethnic identity and among the most political (Liu 2015, 2017; Marquardt 2018a, 2018b). As a “costless” signal, it helps co-ethnics identify each other relatively effortlessly, an especially useful characteristic for low-information environments where other ethnic markers, such as a shared homeland or cultural values, would require engagement between parties to demarcate group boundaries (Chandra 2004). Experimental findings demonstrate that the use of ethnic languages can have a significant impact on respondents, ranging from identification of co-ethnics to enhancing trust to exacerbating stereotypes and ethnic divisions (Chang and Lu 2014; Habyarimana et al. 2009; Hu 2020; Perez and Tavits 2019; Ricks 2020).

In short, we know that language matters. The question, then, is about degree. What impact does the use of an ethnic tongue have on audience opinions? And does that effect vary across contexts? Anecdotally and intuitively, we suspect that there is a positive impact. When listeners hear their ethnic tongue, they should feel a higher degree of kinship with the speaker, which should, in turn, encourage a higher level of political support. For instance, Filipino President Rodrigo Duterte is well-known for speaking in Bisaya to cultivate a folksy appeal that enhances his populist approach to politics (see Escalona 2018). Other Southeast Asian examples include Thaksin Shinawatra, former prime minister of Thailand, who released a video on December 16, 2020 via Facebook where he spoke Khammuang, the language native to northern Thailand, to support a candidate in a local election, and Prabowo Subianto, Indonesia's repeat presidential candidate, who applied Javanese language and ethnic symbolism during his campaigns (see also Selway 2021; Croft-Cusworth 2014).¹ The challenge, though, is to determine the degree to which ethnic tongues are effective relative to speaking in the common political language.

Utilizing an online survey experiment, I test the impact of an ethnic tongue against that of the common political tongue in Indonesia, the Philippines, and Thailand. Furthermore, the tests capture variation in the status of ethnic communities across these countries, allowing me to identify the effect of ethnic languages on (1) a group who speaks a peripheral language compared with the central language: Isan (Lao)² versus Standard Thai in Thailand, (2) a politically dominant group with a unique ethnic tongue contrasted against a national tongue: Javanese versus Indonesian in Indonesia, and (3) a politically dominant group whose language is one of two official state languages: Tagalog versus English in the Philippines. The results show that the ethnic tongues have significant impact on listener opinions among both the Javanese people in Indonesia and the Isan people in Thailand but have little impact among the Tagalog people in the Philippines. These findings highlight the importance of ethnic language in shaping political opinion while also showing that such ethnic cues are less important when the group is the linguistic hegemon.

First, though, I briefly review the literature of the impact of ethnic language on politics, including setting out some theoretical expectations. I then discuss my research design and experimental methods. This is followed by a presentation of the results. Finally, I discuss the results and conclude the article with some of the implications of these findings.

Language and Ethnic Identity

Ethnic identification is associated with many politically important behaviors. These range from ethnic conflict and violence (see Horowitz 1985; Varshney 2002), to the creation of government policy and distribution of public goods (see Alesina, Baqir, and Easterly 1999; Selway 2015; Tajima, Samphantharak, and Ostwald 2018), to more mundane examples, such as whether to trust a bureaucrat (Hu 2020). The general finding has been that co-ethnics identify more closely with each other and that their shared identity translates into more within-group prosocial behavior (Habyarimana et al. 2009; Ricks 2020). Social scientists have studied cases of this around the world, including in Africa (Adida 2015; Ferree 2006; Posner 2005), Asia (Barter 2020; Chandra 200; Sumaktoyo 2021), Europe (Brubaker et al. 2006), Latin America (Madrid 2012), and North America (Flores and Coppock 2018). In short, people tend to demonstrate affinity for co-ethnics and support them when given the choice between an identifiable co-ethnic and an alternative individual.

Language is an important cue in mobilizing ethnic identities. Chandra (2004; 2006) argues that language serves as a descent-based attribute that is highly visible and relatively “costless.” In other words, ethnic identification largely relies on attributes that can help determine group boundaries, and the identifiers that are most visible provide rapid information to others about membership within groups. The work of forming an ethnic community is borne by these signals, like phenotype, dress, and language, which allow for co-ethnics to quickly identify one another (Liu and Ricks 2022).³ Language is particularly useful as a signal of ethnic identity because of its high degree of “stickiness” (Chandra 2006). Adopting a new tongue is laborious, especially an ethnic tongue that might not be readily taught in schools, and native fluency is difficult to achieve, particularly later in life. This means linguistic cues can provide important signals to listeners as they seek to distinguish who is part of their in-group and who is not. Because fluency in a tongue is relatively costless to determine and it is ubiquitous in communication, language serves as one of the most effective markers that allows for co-ethnics to identify one another (Habyarimana et al. 2009, 54–56).

With co-ethnicity or kinship having such a strong impact on political behavior, and language providing one of the most important cues for kinship, it is natural to hypothesize that language with which political candidates communicate can shape political opinions (Flores and Coppock 2018). In short, we should see greater levels of political support for those who signal co-ethnicity through linguistic cues. In other words:

Hypothesis 1: All other things being equal, individuals should feel higher levels of kinship for politicians who employ a shared ethnic language.

Hypothesis 2: All other things being equal, individuals who are exposed to politicians speaking in a shared ethnic language should also express higher levels of political support for the speaker.

That said, language's impact could be conditional upon the ethnic group's status in society. Psychology studies have found that ethnic identity can be situational, wherein minorities or marginalized groups tend to have a higher level of ethnic identification, while majority or dominant groups tend to downplay ethnic identification (Phinney and Alipuria 1990; Xu, Farver, and Pauker 2015). Thus, we might expect that the impact of co-ethnic language use to be highest among ethnic groups that are marginalized. Among those who are the dominant or hegemonic groups, we might expect there to be less impact of co-ethnic signals. Thus:

Hypothesis 3: Co-ethnic language should have a higher impact among marginalized ethnic groups than among dominant ethnic groups.

Such hypotheses warrant testing.

Literature testing co-ethnic identification, though, tends to conflate language, appearance, and other markers, such as names, religion, and race (e.g. Adida 2015; Dunning 2010; Habyarimana et al. 2009; Harris and Findley 2014). Even work specific to language can confound markers. For instance, in Flores and Coppock's (2018) experiment testing the impact of Spanish language advertisements in American politics, respondents viewed Jeb Bush, Filemon Vela, or Mike Coffman speaking either English or Spanish in identical advertisements. While using the advertisements allowed for them to control for the effect of Vela, a Latino female, and Bush and Coffman, both white males, the visual cue as well as the prominence of the candidates provided clear additional signals of ethnic identity. Rarely is language isolated for testing. Also, as Hu (2020, 704) argues, distinguishing attitudes about language from other variables is somewhat difficult, and this has been a repeated challenge in research designs testing for the effect of language on attitudes. To address this challenge, Ricks (2020) employed a survey experiment in Thailand based on audio cues alone to gauge the effect of both linguistic register and ethnic tongues on respondent opinions, finding that an ethnic language had a significant impact, even in the Thai context where ethnic identity is not considered to be a major political cleavage (see Selway 2015, 94–95; Ricks 2019a). Building upon this approach, I emulate Ricks's survey design utilizing only audio treatments to test for the impact of ethnic tongues.

Research Design

To test the impact of language on political opinion, I utilize a simple survey experiment across three countries: Indonesia, Philippines, and Thailand. The experiment tested the impact of the dominant political tongue of the country (Indonesian, Filipino English, and Thai, respectively), against the native language of one of the country's significant ethnic groups (Javanese, Tagalog, and Isan, respectively).

I selected these countries and ethnic groups based on multiple criteria. First, all three countries have recent democratic experiences, including elections in 2019, although the quality of their democracies has been questioned (See Arugay 2019; Aspinall and Berenschot 2019; Davidson 2018; Dressell and Bonoan 2019; Ricks 2019b; Sinpeng 2021). Importantly, for the purpose of the experiment, the respondents would have some experience evaluating politicians through political communication (i.e. speeches).

Second, the countries also have clear distinctions between the dominant language of politics and the native language of important ethnic groups. In Indonesia, the national tongue is Indonesian, but over 40 percent of the population are ethnically Javanese (over 95 million in the 2010 census).⁴ Indonesian is based upon Malay, and it was adopted as the country's national language during the independence movement (Sneddon 2003). Malay had been historically a widely used trading language, but it was the native tongue of only a small proportion of the newly developing nation.⁵ Javanese, though, was spoken by approximately half of the population at independence who were concentrated largely on the island of Java but also scattered throughout the archipelago. Javanese is distinct from Malay and the two languages are not mutually intelligible. Javanese has multiple registers, which embeds a social hierarchy wherein higher levels of the language are reserved for use among the elite (Anderson 1990). These distinctions make it quite exclusive, as the language is difficult to learn for non-native speakers, so much so that nationalist leaders decided it was impractical as a national tongue (Anderson 1990; Liu 2015). Indonesian (Malay) was seen as much more “democratic” (Montolalu and Suryadinata 2007). Suffice it to say, Indonesian and Javanese are separate languages.

In the Philippines, the colonial tongue of English remains common parlance in national government and an official language, despite the adoption of Filipino, a slightly modified version of Tagalog, as the national language. Tagalog is of Austronesian derivation and is the native tongue of about 22.5 million people (about 24.4 percent of the population) who live across large portions of Luzon and the area around Manila. English was imported to the Philippines by the American colonial power, and by the end of colonialism the language was widespread enough that it was retained in education and government, although it was most used among the upper classes (Bresnahan 1979). Today it is spoken widely. In recent years, government sources, such as the Department of Tourism, proclaim that “The Philippines is currently the third-largest English speaking country in the world.”⁶ While both English and Tagalog are now in positions of linguistic power, English remains the dominant tongue in politics (Rafael 2016; Tupas 2020). English words have increasingly been integrated into Filipino and Filipino English has taken on a local character, but the two languages are entirely distinct and mutually unintelligible.

And in Thailand, standard Thai is the official tongue and is used in politics, but the Isan people of northeastern Thailand make up almost 30 percent of the country's population. They are ethnically Lao and speak a version of Lao at home, referred to in Thailand as Isan language, and maintain a distinctive regional identity (Alexander and McCargo 2014). Language is the primary marker that distinguishes Isan people from other Thai citizens (see Liu and Ricks 2022; Ricks forthcoming). While Isan (Lao) and Thai belong to the same language family, Tai-Kadai (see Diller, Edmondson, and Luo 2008), they are distinct (Eberhard, Simons, and Fennig 2021; Smalley 1994). The differences between the two may not be as stark as English and Tagalog or Javanese and Indonesian (Enfield 2002), but they are mutually unintelligible to native speakers who had no exposure to the other tongue. Most speakers of Isan can speak and understand standard Thai due exposure through the Thai education system as well as to television, radio, and other media, but native central Thai speakers would be unable to comprehend Isan without some assistance. Indeed, when Isan speakers are portrayed in Thai media, their speech is generally accompanied by Thai subtitles.

Third, the political and linguistic status of each of these groups varies. The Javanese are Indonesia's hegemonic ethnic group. This has been the case since independence. The Javanese have historically dominated national politics, with every president since independence coming from the ethnic group (Bertrand 2004).⁷ Beyond presidents, Javanese are geographically concentrated on the island of Java, the center of political and economic power in the country. While some research has found that the Javanese ethnic identity is slowly declining, Javanese influence permeates Indonesian politics, culture, and society (Ananta, Utami, and Purbowati 2016; Anderson 1990).

The Tagalog in the Philippines dominate much of Mindoro and Luzon, including the capital Manila. The Tagalog are also the country's largest ethnic group (see Liu and Ricks 2022, 3–9).⁸ Tagalog was also chosen as the country's official language in the 1930s, although it was referred to as Pilipino by the Department of Education in 1959. While subtle distinctions exist between Tagalog and Pilipino, for the most part, they are the same tongue, which gave rise to the “language wars” in the 1960s as other groups resisted imposition of Tagalog (Gonzalez 2000). The 1973 Constitution called for the legislature to establish a new language named Filipino as a common national language. Filipino was based upon Pilipino (Tagalog), reflecting the preferences of Tagalog-speaking political leaders (Tupas 2015). Conflict over the use of Filipino (Tagalog) has led to periodic political battles, including legislative walkouts by non-Tagalog representatives, but acceptance of the language has grown over time (Gonzalez 2000). Geographic location, political dominance, and linguistic hegemony thus place Tagalog speakers as a central ethnicity in the Philippines (Gonzalez 1980).⁹

In contrast, Isan people are politically, economically, and geographically marginalized in Thailand, despite composing almost one-third of the country's population (Draper and Selway 2019). The region has been left behind in the Thai state's efforts to develop Bangkok, and the Northeast region where Isan people are concentrated ranks relatively poorly in education, economic development, income levels, and health outcomes (UNDP 2014). The Isan people have historically been excluded from political power, sometimes through violent means (Keyes 2014). A regional identity has formed (McCargo and Hongladarom 2004; Alexander and McCargo 2014), and the region gained some political prominence under the administration of Thaksin Shinawatra (2001–2006), but ethno-regional mobilization remains minimal (Ricks 2019a).

Thus, the Javanese are the geographically and politically dominant ethnic group in Indonesia, but their language is distinct from the national tongue. The Tagalog are also dominant thanks to geography and language; even so, the Philippines still employs the colonial tongue in politics. And the Isan people, despite their numbers, are marginalized in Thailand, geographically, politically, and linguistically.

Distinctions between mother tongues and the politically dominant language in all three cases allow me to test Hypothesis 1 and Hypothesis 2, while variation in the status of mother tongues allows me to test Hypothesis 3. Finally, beyond capturing variation in languages and status, the size of these groups also gave me a sufficiently accessible target population to sample using an online platform.

The experiment was embedded in an online survey conducted in May 2020, and the design was developed to allow for comparison with the results of Ricks's (2020) face-to-face survey findings regarding Isan language. The survey was self-administered via handphone, tablet, or computer, and allowed for respondents to withdraw from the survey at any time. Overall, the entire survey was designed to take approximately 20 minutes to complete. Respondents were recruited via the Qualtrics survey sample, and they were offered a small remuneration for their participation through awards such as coupons. Only native speakers of the ethnic language in question (Javanese, Tagalog, or Isan) were recruited into the experiment.

The survey treatment was a short excerpt of a political speech, recorded in either the common political tongue (Indonesian, Filipino English, or Thai) or in the dialect of their ethnic group (Javanese, Tagalog, or Isan). The content of the speech was translated from the same base script¹⁰ and included an innocuous appeal to national sentiments, and in each of the three countries both forms of the speech were recorded by the same voice actor, an adult male member of the ethnic group, to preempt any preference bias based on the voice. The recordings were approximately two minutes in length, depending on speaking speed and language. Assignment of the treatment was random, with the online platform designed to expose approximately half the sample to the speech in their ethnic language and the other half to the speech in the politically dominant language.

Respondents first answered a series of demographic questions as well as some opinion questions before being asked to listen to the recording. They were told that the recording was of a political speech. No other visual or written cues about the speaker were provided. After hearing the treatment, respondents were presented with a series of 14 statements about the appeal of the speaker (presented in Tables 3 and 4, below) and were asked to mark the degree to which they agreed with each statement on a five-point Likert scale ranging from strongly agree to strongly disagree. Statement order was randomized to minimize any impact of question sequence on responses. The 14 statements were chosen to capture respondent feelings across three categories: (1) statements that suggest electoral support, (2) statements that suggest kinship, and (3) statements that reflect fitness for office. The statements in the kinship category were designed to test for hypothesis 1, while statements in the electoral support category were designed to test for hypothesis 2. The fitness for office category was included to maintain consistency with Ricks (2020), and to gauge whether the use of an ethnic tongue shaped perceptions about whether a politician might be less prepared for national office than one who spoke the common political language. Ricks (2020) had found Isan language to have a mixed impact on this category, so I was agnostic as to the category's potential outcome.

After data collection, the data was cleaned, excluding straight-line respondents as well as those who had completed the survey too quickly or had taken inordinately long. The final set of respondents totaled 1,252, split relatively evenly across the three countries (Indonesia: 444; Philippines: 408; Thailand: 400). In each country, approximately half the sample heard the dominant political language while the other half heard their native tongue.

I must here acknowledge the caveat that the samples were not random, due to respondents self-selecting as survey participants. As such, their demographic composition may not completely reflect the larger population. Nevertheless, the descriptive statistics suggest that the sample provided sufficient variety to provide confidence in the findings (Mullinix et al. 2015). Also, as reported below, the findings in Thailand were compared with an analysis of replication data from Ricks's (2020) face-to-face survey experiment, showing a remarkable degree of consistency, giving increased confidence in validity of the experimental results.

Results

Following the discussion above, I expected that the treatment groups who heard the speech recorded in their native ethnic language would rank the speaker more favorably in terms of kinship and electoral support than those who listened to the speaker in the common political language. I also expected, though, that this effect should be stronger among the marginalized Isan people in Thailand than among the Javanese in Indonesia or the Tagalog in the Philippines.

To get a rough sense of responses, we can look at an extract of the raw data on two of the statements, as reported in Table 1. The data shows the distribution of responses varies a fair amount between those who heard Javanese and Indonesian in Indonesia as well as Isan and Thai in Thailand. For example, over 86 percent of Thai respondents who heard Isan agreed or strongly agreed with the statement, “The speaker was likely born in the same region as me”; in contrast, only about 47 percent of those who heard the Thai version felt the same. The respective numbers are 77 percent and 42 percent for Javanese versus Indonesian in the Indonesia sample. The differences between the Tagalog and English treatment groups in the Philippines appear much smaller. In fact, they are negligible. On the same statement, almost 68.5 percent of those who heard Tagalog and 68.1 percent of those who heard English agreed or strongly agreed.

Table 1. Distribution of responses across two statements

Note: Numbers presented are percentages with the actual number of respondents in parentheses for each category.

Variation in responses regarding a willingness to vote for the speaker were less pronounced but still large in Thailand, with about 70 percent of those who heard Isan agreeing or strongly agreeing that they would consider voting for the speaker while only 56 percent of those who heard Thai said the same. In Indonesia, 53 percent of the Javanese treatment group felt the same versus about 42 percent of the Indonesian treatment group. Tagalog and English saw only minimal differences on the same standard, about 62.7 percent of the Tagalog treatment group versus 62.8 of the English treatment group. These numbers provide some initial support for the hypotheses, although it appears that the Tagalog people in the Philippines are not influenced by their native tongue.

While this raw data does suggest that we see an impact from language in Thailand and Indonesia but not in the Philippines, we can subject this to more rigorous testing. Initially, I tested for covariate balance between treatment groups (ethnic language) and control groups (common political tongue) across a set of descriptive variables including age, sex, income levels, education levels, and the respondents’ living circumstances (whether they lived in a city, suburbs, or a rural community).¹¹ The Hotelling statistic (reported in Table 2) allows us to test whether the two groups are substantially different across this group of variables. The null hypothesis is that the two groups have the same multivariate mean; in other words, the two groups are comparable. A p value of less than 0.1 would potentially suggest that the two groups are substantially different based on this combination of variables, and thus give us pause as to whether our treatments are sufficiently randomized. As we see here, though, all three tests suggest the groups are commensurate.

Table 2. Covariate Balance across Treatment Groups

Note: Hotelling statistic is calculated on five variables: Age, Sex, Income categories, Education level, and an Urban-Rural variable. Ricks (2020) reports Hotelling statistics in table 2 (p. 95).

After confirming that the treatment groups were comparable, I tested my hypotheses regarding the impact of an ethnic tongue versus the common political tongue using difference of means analysis, with the control group serving as the base category. Responses were ordered on a five-point scale with a more positive response being associated with a higher number. Results are reported in Table 3 below, including an analysis of replication data from Ricks (2020) for comparison between the online survey results and the face-to-face survey conducted in 2016. For ease of interpretation, though, the results of the online survey are also presented in graphic format in the discussion section below (Figures 1 through 3).

Figure 1. Treatment Effects of Isan vs Thai in Thailand.

Note: Dots indicate difference of means results with the Thai treatment group serving as the base category. Bars indicate 95% confidence intervals. Positive numbers indicate that respondents who heard the Isan language treatment ranked the speaker more favorably than those who heard the Thai language treatment, while negative numbers indicate the reverse. See Table 3 for exact point estimates.

Table 3. Treatment Effects of Ethnic Language over Common Political Language

Note: Numbers report the difference of means between responses from the group that heard the statement in their native tongue and the group that heard the statement in the common political language as indicated. Standard errors are in parentheses. Data in the final column is calculated using replication data from Ricks (2020). * p < 0.1, ** p < 0.05

In short, difference of means tests show a similar pattern to that which we saw in the raw data. Ethnic language appears to have a substantial positive impact on both the Isan group in Thailand and the Javanese group in Indonesia. In the Philippines, though, we see almost no impact from the difference between Tagalog and English on respondent opinions. I discuss the findings in more detail in the discussion section below. Importantly, we see that the results from the Isan experiment and data from Ricks (2020) are largely consistent despite the two experiments being conducted via different means (online versus face-to-face) and four years apart. This gives us greater confidence in the results of both surveys.

To further gauge the degree of impact language has on respondent opinions, I also conducted a marginal effects analysis on all three experiment cohorts as well as replication data obtained from Ricks (2020). This also allows for a secondary check on the efficacy of treatment randomization, as I can include a set of control variables in the analysis. To do this, I truncated all responses to the set of 14 statements to produce binary variables where a 1 represents either “strongly agree” or “agree” while a 0 represented all other marks. This results in dependent variables that capture only clearly positive responses.

Utilizing these new binary dependent variables, I employed a logit model with control variables of age, sex, income, education levels, fixed effects on living conditions (rural vs urban), and fixed effects of the respondent's birth province:¹²

$$Y( {binary variables} ) = \beta 0 + \beta 1Xtreatment + \beta 2Xage + \beta 3Xincome + \beta 4Xsex + \beta 5Xeducation + \beta 6X( {i.hometype} ) + \beta 8iX( {i.provbirth} ) + \varepsilon $$

From this, I obtained the marginal effect of hearing the ethnic language over the common political tongue. These results are reported in Table 4. In general, the pattern seen in the difference of means analysis holds. In the online survey experiment, both Javanese and Isan are seen to have a substantial impact on the probability of a respondent to “strongly agree” or “agree” to many of the statements, with the strongest impact being see in the kinship category. Indeed, an Isan person who heard the Isan language clip was 37.5 percent more likely to mark either of those categories than if they heard the Thai language clip. The Javanese who heard the Javanese clip were over 30 percent more likely to do the same.

Table 4. Marginal Effects of hearing the Ethnic Language over Common Political Language

Note: Average marginal effect of hearing a statement in native language over hearing it in the common political language. Standard errors in parentheses. Data in the final column is calculated using replication data from Ricks (2020).

* p < 0.1, ** p < 0.05

The effects are also seen in the electoral support category, where Isan language improves a respondent's chances of expressing higher levels of support in the political sphere by between 13 and 19 percent in all categories. The impact of Javanese was slightly smaller, but still significant, running between 9 and 12 percent. The only exception where the ethnic tongue did not have an impact was on the statement about local government, likely due to compression of the dependent variable.

The fitness for office category saw the largest change from the findings from the difference of means tests, mostly due to the compression of the data into a dummy variable, meaning we lost some of the leverage available in the Likert scale. In Indonesia, only the statement on persuasiveness of the speaker saw a significant difference where difference of means tests had seen significance across the board. In Thailand, the outcomes were the same as those reported in the difference of means analysis.

The results from Ricks's (2020) data also remain the same except in three categories. In two of them (national representative and local government), the difference of means effects were already somewhat small, and so it is not surprising that the impact disappears when the data is truncated into a binary variable. The biggest difference is in the statement “The speaker would likely be able to represent my interests in policymaking,” which saw a relatively large difference of means, while the effect has disappeared in the marginal effects analysis. This, again, is due to the nature of responses, where the Likert scale saw those who heard Isan much more likely to strongly agree with the statement, while those who heard Thai were much more likely to disagree or strongly disagree.

In the Philippines, we again see that Tagalog versus English does not have much of an effect, largely in agreement with the numbers reported in the main essay. Tagalog language does have an impact in terms of perception of social class. We also see, though, some impact on two statements that we did not see in the difference of means analysis. This was again due to the truncating of the data.

Overall, the marginal effects analysis corroborates the findings found in the difference of means tests, giving us confidence that the treatments were sufficiently randomized. Because of collapsing the data into binary variables, though, there are some slight differences. Marginal effects analysis, while useful in communicating results in substantive terms, does lose some of our empirical leverage from Likert scales.

Discussion

Based on my hypotheses, I expected that the ethnic language should have an impact across all three of my respondent groups, but the impact should be most pronounced among the marginalized Isan community in Thailand. The Javanese and the Tagalog should see a smaller effect based on the two groups’ dominant status in their perspective countries. Following expectations, we do see language having considerable impact among the Isan people of Thailand. Among the Tagalog of the Philippines, though, we see almost no effect, which could be interpreted in line with expectations based on the hegemonic position of the Tagalog language. Even so, I did expect that language would have at least some influence, so the lack of results deserves discussion. Among the Javanese people in Indonesia, though, we see something of a surprise. Language has a strong effect. Almost on par with that which we see in Thailand. I discuss the results in that order.

First, the Isan experiment fits relatively well with our theoretical expectations. In both the online survey and the replication data from Ricks (2020), hearing a speaker use Isan language has a substantial impression on respondent opinions toward the speaker across almost all 14 statements presented them. Figure 1 provides a graphic representation of these results, showing that the linguistic treatment had significantly positive effect on respondent opinions for all but two of the statements. The strongest results were seen in the kinship category, while the weakest were in the fitness for office category.

Perceptions of kinship with the speaker were much higher among those who heard the Isan language clip. In terms of degree, respondents were over 37 percent more likely to agree or strongly agree that the speaker came from their home region and 30 percent more likely to agree that he shared a similar background to their own when compared with group who heard the same speech in Thai. Respondents in the Isan treatment group were also more likely to think the speaker shared their opinions, understood their challenges, and came from the same social class. The language was clearly acting as an identity signal for those who heard it.

This effect carries onto statements of political support. Those who heard the Isan speech were approximately 14 percent more likely to choose “strongly agree” or “agree” when asked whether they would consider voting for the speaker than those who heard the statement in central Thai, signifying the substantial impact of language on respondent opinion. They were also more likely to view the speaker as a good candidate for either regional or national office, and they were over 15 percent more likely to strongly agree or agree that they could trust the speaker to represent their hometown. This appears to be driven by the impact of feelings of kinship, as we see some variation in respondent feelings regarding fitness for office based on language. Respondents who heard Isan did not rank the speaker noticeably higher in terms of education levels or preparedness for office, but they did provide him substantially higher marks in terms of kinship indicators.

We should note, as well, that results from this experiment largely correspond to those of Ricks (2020). The only significant difference between the difference of means results is found in the statement on social class, wherein the effect was more pronounced in our online sample. This could potentially be explained in the demographics of the online sample, which were somewhat more affluent than those in the face-to-face survey. The similarity of the findings, drawn from two independent samples four years apart, should give us increased confidence in the conclusions drawn from both experiments.

Second, the experiment in the Philippines, in contrast to Thailand, showed that respondents did not rank the speaker significantly differently based on which of the two language treatments they heard, as shown in Figure 2. The only substantial difference was on the statement regarding social class, wherein those who heard Tagalog ranked the speaker as more likely to share their class status. Across all other statements, we see no statistically significant difference brought on by exposure to Tagalog versus English.

Figure 2. Treatment Effects of Tagalog vs English in the Philippines.

Note: Dots indicate difference of means results with the English treatment group serving as the base category. Bars indicate 95% confidence intervals. Positive numbers indicate that respondents who heard the Tagalog language treatment ranked the speaker more favorably than those who heard the English language treatment, while negative numbers indicate the reverse. See Table 3 for exact point estimates.

While I was expecting the impact of language among the Tagalog people to be lower than among Isan people, I was not expecting it to be practically non-existent inasmuch as some scholars suggest that there is a tension between English and Tagalog use in the Philippines wherein some Tagalog speakers might resent the use of the colonial language. The prevalence of English in government means that it is frequently treated as the more “official” of the official languages, with Tupas (2020, 232) stating that “English language reigns supreme … because it is the most powerful, both symbolically and materially” (see also Borlongan 2009). Indeed, past national politicians have used English to demonstrate their capacity as leaders. This has resulted in some resentment toward English and calls for more use of Tagalog, with English seen as a colonial leftover (Rafael 2016; Tinio 2009). Nevertheless, we see no clear evidence of this tension in the results.

One explanation for this departure from my expectations could be that both English and Tagalog (Filipino) are treated as official languages in the Philippines, giving both a similar appeal. Being granted official status as well as hegemonic status over the other languages in the archipelago, Tagalog has become less identifiable as a symbol of ethnicity. Indeed, other ethnics within the Philippines are schooled in both Filipino (Tagalog) and English, reducing their exclusiveness. Thus, the comparison between the two may provide less leverage than seen in Thailand and Indonesia.

Additionally, ethnicity in the Philippines is relatively apolitical, apart from the Moro group in the south. Despite multiple ethnic cleavages and groups, ethnic differences have not become the subject of political divisions or party mobilizations,¹³ and some scholars lump most ethnic groups in the Philippines together in a low-land Catholic mass (May 2003). Thus, ethnic identification may play only a minor role, if any, among Tagalog people.

That said, there was one statement where Tagalog did have an impact: social class. Respondents were more likely to identify the speaker as someone from their social class when he used Tagalog. This suggests that English may signify being part of the upper class, but this does not significantly distance a speaker from voters. Is this result due to our sample coming from the upper classes? First, our respondents reported their income ranges, with 76.4 percent of the sample reporting that their monthly income was below 20,000 pesos per month and only 3.69 percent (15 respondents) reporting an income over 40,000 pesos per month. In 2015, the Family Income and Expenditure Survey reported that average household income was about 22,000 pesos per month, meaning that our sample was not overtly skewed toward higher social classes. Second, to gauge whether this might be an artifact of our sample coming largely from Manila (197 of 408 respondents or 48.3 percent), I repeated the analysis excluding Manila-based respondents. The results remained largely the same in terms of statistical significance.¹⁴

As such, the data suggest that among Tagalog people in the Philippines, the impact of Tagalog versus English is not sufficiently strong as to privilege one language over the other. A politician would likely accrue the same benefit no matter which of the languages she decided to speak. Of course, this finding may be limited to only the Tagalog ethnic group. Anecdotal evidence suggests that President Rodrigo Duterte has increased his home-town appeal by mixing in Bisaya language during his speeches (Escalona 2018). The question then is whether his charm is based in language or in style, as his speech patterns are purposefully anti-elite; this is a challenge for further testing. At the very least, the knowledge that there is no significant difference between respondents’ reactions to English and Tagalog suggests that the Tagalog language is not a significant ethnic identifier in the Philippines.

Third, among Javanese respondents, hearing a speaker use Javanese results in statistically significant differences in opinions regarding the speaker across the board, as seen in Figure 3. Indeed, the impact of the Javanese language among the Javanese when measured by difference of means tests appears to be just as strong as Isan language among Isan people. In a small contrast to Thailand, there are no categories where the ethnic tongue does not have a measurable positive impact on respondent opinions. Marginal effects, though, are not as high, thanks to collapsing data into binary variables.

Figure 3. Treatment Effects of Javanese vs Indonesian in Indonesia.

Note: Dots indicate difference of means results with the Indonesian treatment group serving as the base category. Bars indicate 95% confidence intervals. Positive numbers indicate that respondents who heard the Javanese language treatment ranked the speaker more favorably than those who heard the Indonesian language treatment, while negative numbers indicate the reverse. See Table 3 for exact point estimates

Still, though, in most cases these remain significant. For instance, using marginal effects analysis, we see results from the statement “I would consider voting for the speaker if he were to run in an election” show that respondents who heard the Javanese speech were 11.3 percent more likely to either strongly agree or agree with the statement than those who heard the Indonesian version of the speech. This degree of difference, if translated into action at the voting booth, could potentially change the outcome of many electoral contests. In other words, language matters in shaping Javanese voter opinions.

We can see some hints as to the cause of the variation across the statements. Those statements related to feelings of kinship were among the highest ranked among respondents in the treatment group, suggesting that the use of Javanese language established higher feelings of closeness and trust in the minds of the respondents, like our results among Isan people. Respondents were more likely to believe the speaker was born in the same region, had a similar background, and was in the same social class as themselves. Indeed, respondents who heard the speech in Javanese were over 30 percent more likely to agree that the speaker came from their home region and over 18 percent more likely to agree that he shared a similar background to themselves than those who heard the same person speak in Indonesian. These feelings potentially contributed to the perception that they could trust the speaker and support him as a political leader.

As noted earlier, I had expected that Javanese language would have an impact, but I was expecting it to be weaker than that seen in the peripheral Isan group in Thailand. Such was not the case. This raises the question as to why a dominant political group would still respond so favorably to their ethnic language despite already being so prevalent in the society. Has integration of the Indonesian language failed to reduce ethnic identification?

The data presented here has no clear answers, but we do know that Javanese remains a vibrant language, with over 68 million regular users, making it, by far, the largest linguistic group in Indonesia (Saddhono and Rohmadi 2014; Eberhard, Simons, and Charles D. Fennig 2021).¹⁵ Even so, there are increasing pressures in the country for Javanese people to adopt alternative languages, especially Indonesian (Cohn and Ravindranath 2014; Smith-Hefner 2009). For some Javanese, this could be seen as a threat. To employ the words of Anderson (1990, 235), who was writing about the Sundanese, the Javanese today “may fear that their language is … in danger: danger of dying out by neglect, danger of becoming irrelevant, danger of being crowded out by Indonesian.”¹⁶ Such threat might push them to cling more tightly to political leaders who could speak the ethnic tongue.

Another possible explanation might be that, even though the Javanese are the hegemonic ethnic group in Indonesia, their dominance is not so complete as to allow the group to feel at ease with the national tongue. Bazzi and colleagues (2019) have argued in the Indonesian context that when there are few large groups in a community, or what they call high polarization, relationships between ethnic groups can become increasingly strained. More so than when a series of smaller groups must integrate. Perhaps some Javanese respondents, seeing their ethnic community making up approximately 40 percent of the population, feel as though other ethnic groups derive greater benefit from Indonesian than the Javanese and, therefore, Indonesian may be a potential contender against their ethnicity and identity.

Or perhaps there is an ethnic pride among the Javanese built upon centuries of Javanese tradition, courtly behavior, literature, and language, which may have something to do with Javanese political behavior (Anderson 1990; Pemberton 1994). We do know that the Javanese flocked to support Joko Widodo, a Javanese co-ethnic, in the 2019 presidential election, signaling an ethnic cleavage between the Javanese and non-Javanese (Pepinsky 2019).¹⁷ The use of Javanese language could be an important signaling device for political mobilization among the Javanese.

Of course, the data presented in this article does not provide us clear direction as to which of these explanations might be most correct, or if all or none of them are, but these results are suggestive that the nation-building project of Indonesia is on-going, and it is important to recognize potential social cleavages that might derail the process, such as an over-emphasis on ethnic identity (see also Sumaktoyo 2021). As Indonesia has long dealt with ethnic tensions (Bertrand 2004; Davidson 2008), we must acknowledge that the dominant Javanese ethnic group could be susceptible to mobilization through ethnic symbols.

Lessons drawn from these data are subject to additional testing, as an online sample should not be taken as definitive alone (Mullinix et al. 2015). Also, all three of these countries are home to multiple ethnic groups, and as we know from our findings, there is significant variation in the degree to which different groups respond to ethnic tongues.

In conclusion, the results provide contributions to both the study of Southeast Asia and the study of language politics. First, we see potential political power of ethnic languages in Southeast Asia. In both Indonesia and Thailand, ethnic language had a considerable impact on respondent's willingness to consider political support for the speaker. Despite their hegemonic position in Indonesia, we see that the Javanese could still be mobilized by ethnic identity (Anderson 1990; Pepinsky 2019). And despite Thailand's long-standing effort to present itself as ethnically homogenous (Streckfuss 2015; Ricks 2019a), we see that ethnic language remains capable of shaping people's behavior, suggesting the resilience of ethnic identities (Alexander 2021; Selway 2020, 2021). National tongues are designed to unify (Liu 2015), but such policy efforts can be undermined when ethnic cleavages become mobilized in political fights (Selway 2015). The nation-building efforts of Southeast Asian states are ongoing, and language will continue to be a central component of the process.

Second, with both the Javanese and Tagalog being dominant ethnic groups, the variation we see in their results provides interesting insight. The lack of any substantial results between English and Tagalog suggests Tagalog plays little role in ethnic identification, while among the hegemonic Javanese it does. This hints that claims over the tension between Tagalog and English (Rafael 2016; Tinio 2009) are perhaps overstated. It also implies that the elevation of Tagalog to the status of both national and official language has reduced its capacity to act as an ethnic identifier relative to the other official language of English. Javanese, though, is excluded from national language status. Liu (2015, 53–57, 61–62) argues that this was a power-neutralizing act, but the fact that Javanese people can still be mobilized by the tongue indicates that cultural egoism may still be relevant.¹⁸ Javanese continues to play an important role in ethnic boundary-making. In short, this comparison suggests that giving a language official status may have an important role in determining its capacity in ethnic identification, or the lack thereof.

Finally, for the study of language politics, the results of this experiment confirm that even short exposure to an ethnic tongue, independent of other identifying markers, can significantly enhance feelings of kinship between a listener and a speaker. In terms of degree, these impacts can be considerable, as demonstrated in the marginal effects analysis which showed both Javanese and Isan people were over 10 percent more likely to express political support and over 30 percent more likely to express regional identification with the speaker when they heard their ethnic tongues. As Chandra (2004) notes, language is one of the “costless” signals of ethnic identification, meaning that it helps individuals identify co-ethnics without expending resources such as learning about their background or history or beliefs. Combining this costlessness with the fact that minimal exposure to language can mobilize identity implies that language is one of the most important indicators of ethnicity (see Marquardt 2018a, 2018b) and reinforces the importance of the study of language in politics.